ML researcher at NJUST with 96 works in 3 years and h-index 56, focused on distributed & stochastic optimization.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
56
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-15) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
The output axis is not scored: cumulative and per-year publication totals are missing from this snapshot, so productivity beyond the recent three-year window cannot be judged. It would measure total and annual publication volume.
Impact—
The impact axis is not scored although raw citation aggregates (h-index 56, i10-index 129) appear in the snapshot; citation distributions and venue-level influence are not yet available, so no estimate is made. It would measure citation-based research influence.
Focus0/100
The radar scores focus at 0, reflecting a strongly dispersed topic profile: OpenAlex's five largest topics are closely spaced — advanced neural network applications (102 works), advanced image and video retrieval (90), image retrieval and classification (77), multimodal machine learning (60), and domain adaptation / few-shot learning (55). No single cluster dominates, and the two Evolve-reviewed 2026 papers add distributed and stochastic optimization, widening rather than narrowing the spread.
Collaboration39/100
Collaboration scores 39, consistent with a compact co-author pool: the snapshot lists six distinct coauthors, with Wei Jiang and Lijun Zhang each appearing twice in the frequent-coauthor list alongside Sifan Yang and Yibo Wang. Both Evolve-reviewed 2026 papers are co-authored optimization works, indicating a small, stable team rather than a broad or rapidly rotating collaboration network.
Activity100/100
Activity earns the maximum score of 100. The snapshot records 96 works in the most recent three years, and annual counts climb from 22 (2024) to 41 (2025), the highest in the series, with 33 already logged in 2026. Earlier cycles — 27 works in 2017 and 20 in 2018 — show this intensity is sustained rather than new.
Reproducibility55/100
Reproducibility scores 55 across two assessed papers (reproMean 55). The sign-based variance reduction study is rated medium but reports no experiments, so its claims rest on theory alone; the FCCO extension also rates medium, with strong experiments yet a single SOX baseline and no variance reporting. Both ratings come from Evolve's reproducibility-readiness review of the arXiv versions.
Research summary
Zechao Li is a researcher at Nanjing University of Science and Technology (China) whose public record is name-resolved from OpenAlex bibliographic data. The snapshot reports an h-index of 56, an i10-index of 129, and 96 works in the most recent three years, with annual output rising from 22 works (2024) to 41 (2025) and 33 already logged in 2026 — the basis for a maximal activity score of 100. OpenAlex topic modeling spreads the record across five large, closely sized clusters: advanced neural network applications (102 works), advanced image and video retrieval (90), image retrieval and classification (77), multimodal machine learning (60), and domain adaptation with few-shot learning (55); this even distribution underlies the focus score of 0. The two papers Evolve has reviewed, both 2026 arXiv preprints, sit on the optimization side of the record: a theory paper showing sign majority voting is biased and that server-side compressed gradient tracking attains centralized-optimal distributed rates, and a journal extension proposing the MSVR estimator for finite-sum coupled compositional optimization. Collaboration runs through a small pool of six distinct coauthors, with Wei Jiang and Lijun Zhang the most frequently repeated. Output and impact axes remain unscored pending citation-level data.
Collaboration pattern
Collaboration is concentrated in a small, stable pool: the snapshot lists six distinct coauthors, with Wei Jiang and Lijun Zhang each appearing twice in the frequent-coauthor list, alongside Sifan Yang and Yibo Wang. Both Evolve-reviewed 2026 papers are multi-author optimization works, and the radar's collaboration score of 39 reflects this compact network rather than a broad, fast-rotating set of partners.
Research keywords
Highlights
Removing bias from distributed sign optimization
MSVR estimator for finite-sum coupled compositional optimization
Timeline
- 2010Indexed output begins in the snapshot with 3 works in 2010 (a single 1999 record also appears; 2 works follow in 2011).
- 2017First output peak: 27 works, the highest annual count before the 2020s, followed by 20 works in 2018.
- 2021A low point with 8 works, before output rebounds to 19 (2022) and 18 (2023).
- 2025Highest annual count in the snapshot: 41 works, bringing the most recent three-year total to 96.
- 2026Two optimization-theory preprints reviewed by Evolve: distributed sign-based variance reduction, and the MSVR estimator for finite-sum coupled compositional optimization.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometrics are drawn from OpenAlex only; the h-index (56) and i10-index (129) are quoted as reported in the snapshot and not independently verified.
- The radar's output and impact axes are unscored due to insufficient data; no values are estimated for them in this profile.
- The OpenAlex topic profile (retrieval, multimodal learning, few-shot) diverges from the two Evolve-reviewed optimization papers; Evolve's catalog sample here is only two 2026 arXiv preprints, so the fieldDist split (0.5/0.5) reflects that sample alone.
- 2026 counts cover an in-progress year, and the profile is name-resolved from public bibliographic data, so record mixture across same-name authors cannot be excluded.